apr-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ryan Bloom" <...@covalent.net>
Subject RE: [PATCH 2] example binary BUSEC patch for benchmarking only
Date Thu, 11 Jul 2002 16:33:50 GMT
Brian, fix is commit soon.

Ryan

----------------------------------------------
Ryan Bloom                  rbb@covalent.net
645 Howard St.              rbb@apache.org
San Francisco, CA 

> -----Original Message-----
> From: Brian Pane [mailto:bpane@pacbell.net]
> Sent: Thursday, July 11, 2002 9:31 AM
> To: APR Development List
> Subject: RE: [PATCH 2] example binary BUSEC patch for benchmarking
only
> 
> On Thu, 2002-07-11 at 06:58, Bill Stoddard wrote:
> > I ran a quick profile with this patch and it eliminated a couple of
> > divisions (calls to __divi64 reduced from 4 to 2 in my test setup.
your
> > mileage may vary) which was good for 493 instructions. Still have 3
> __divu64
> > and 2 __divi64 calls. The three __divu64 calls are in the
gettimeofday()
> CRT
> > function, so there is not much we can do about these directly.  One
> __divi64
> > is in apr_poll (convert microseconds to milliseconds. This can
probably
> be
> > optimized away). The other __divi64 is somewhere in cached_explode
> > (util_time.c).
> 
> With the cached_explode() division now fixed, I just looked at the
> division in apr_poll().  The code that's there now is doing the math
> like this:
> 
>     if (timeout > 0) {
>         timeout /= 1000; /* convert microseconds to milliseconds */
>     }
> 
> It's assuming that timeout, an apr_interval_time_t, represents an
> absolute time in microseconds--but that may not be true much longer
> if we switch to binary microseconds.  The calculation should be
> something like:
> 
>     timeout_msec = apr_time_usec(timeout) / 1000;
>     timeout_sec = apr_time_sec(timeout);
>     if (timeout_sec) {
>         timeout_msec += 1000 * timeout_sec;
>     }
> 
> I don't see a way to eliminate the "/ 1000" to convert usec to
> msec.  But we may be able to do all the math in 32-bit mode, by
> limiting the maximum timeout to the number of milliseconds that
> will fit in 32 bits, which works out to a max timeout of about
> 50 days.
> 
> --Brian
> 



Mime
View raw message